examples/train/multimodal/omni/infer.sh

CUDA_VISIBLE_DEVICES=0 \
VIDEO_MAX_PIXELS=50176 \
FPS_MAX_FRAMES=12 \
MAX_PIXELS=1003520 \
ENABLE_AUDIO_OUTPUT=0 \
swift infer \
    --adapters output/vx-xxx/checkpoint-xxx \
    --stream true \
    --load_data_args true \
    --max_new_tokens 2048